Essays in environmental and resource economics

(1)

Tilburg University

Essays in environmental and resource economics

Michielsen, T.O.

Publication date:

2013

Document Version

Publisher's PDF, also known as Version of record Link to publication in Tilburg University Research Portal

Citation for published version (APA):

Michielsen, T. O. (2013). Essays in environmental and resource economics. CentER, Center for Economic Research.

General rights

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain

• You may freely distribute the URL identifying the publication in the public portal Take down policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

(2)

Essays in Environmental and Resource Economics

(3)

(4)

Essays in Environmental and Resource Economics

Proefschrift

ter verkrijging van de graad van doctor aan Tilburg Uni-versity, op gezag van de rector magnificus, prof. dr. Ph. Eijlander, in het openbaar te verdedigen ten overstaan van een door het college voor promoties aangewezen commissie in de aula van de Universiteit op woensdag 25 september 2013 om 10.15 uur door

Thomas Olivier Michielsen

(5)

Promotiecommissie:

Promotor: _{prof. dr. Reyer Gerlagh} Overige Leden: _{prof. dr. Aart de Zeeuw}

(6)

Acknowledgements

I would like to thank a number of people for their help in writing this dissertation. First of all, it has been an enormous pleasure to work with my supervisor, Reyer Gerlagh. Reyer was always willing to interrupt his work to listen to my thoughts, ramblings and conjectures, and respond to them with an astounding intuition and accuracy. Often, after deciphering the essence from my unstructured monologues - no mean feat in itself - he would quickly suggest a result or proof technique that as an unbelieving Thomas I initially discredited, only to come around after a hard day (or two) of work. I also appreciate his informality and wit; working with Reyer was not only learnful, it was also fun. PhD students at other universities have repeatedly told me how lucky I have been to have Reyer as a supervisor, and I could not agree more.

Secondly, I would like to thank Rick van der Ploeg. He has shown incredible kindness and hospitality when inviting me to spend two terms in Oxford, and I enjoyed our conversations about research, academic life and other subjects. I also valued his advice about the job market, as well as his patience with the considerable indecisiveness I have shown in my job search. Thanks for this are also due to Aart de Zeeuw and the members of the job market committee. I mirthfully recall Eline van der Heijden’s response when I told her I had agreed to go to Oxford but had not yet signed the contract. Presciently, she warned me that I might change my mind yet again, although my subsequent change of affiliation from the Department of Economics to New College was not as large a switch as to, say, an energy company.

Thijs ten Raa persuaded me to consider a career in academia and start a PhD. His interesting classes in public economics and industrial organization prompted me to seek him out as a thesis supervisor. I am grateful to him for giving me the confidence that I could complete this four-year trajectory, a commodity with which I have not always been naturally endowed.

Arian has been a great office mate during the last three years. Our conversations were great fun, and I learned a lot from your experiences and backgrounds which were

(7)

Acknowledgements

different from mine. You were always willing to put up with my quirks and listen to my frustrations and successes and my attempts - with varying degrees of success - to explain my research in everyday language. For making my time as a PhD student more enjoyable I would also like to thank the members of my lunch group, whom I hope did not feel underappreciated because of my frequent seminar attendance.

Lastly, I owe thanks to my friends and family and everyone else who has given me strength and has kept reminding me that the world is a more diverse, nuanced and beautiful place than I sometimes tend to think it is.

(8)

Chapter 1 Introduction

Up until 1800, economic growth has been non-existent or modest at best: between the year 1 and 1820, real income per capita growth was only 0.02% per year on average (Stutz, 2010). The spectacular escape from the Malthusian conjecture of low and stagnant per-capita incomes from the Industrial Revolution onwards has been enabled by a shift away from land-intensive to capital-intensive production techniques, particularly in the energy sector (Hansen and Prescott, 2002). The UK’s overreliance on wood as a fuel resulted in serious scarcities in the sixteenth century (Ray, 1979). Though the substitution of coal for wood for domestic heating and industrial use initiated much earlier,1 _{the pressure on}

British wood supplies only abated after the succesful adoption of coke in all stages of the iron smelting process in 1784 (Warde, 2006). Fossil fuels have since offered a cheap, reliable and abundant source of energy. Jevons (1865) extols their significance as follows:

”Coal in truth stands not beside but entirely above all other commodities. It is the material energy of the country - the universal aid - the factor in everything we do. With coal almost any feat is possible or easy; without it we are thrown back into the laborious poverty of early times.”

Jevons was one of the first to warn about the exhaustibility of fossil fuels and the impli-cations of a decline in coal production for living standards. The oil crisis in the seventies of the twentieth century drew renewed attention to the prospect of energy shortages. Optimists trust that the market mechanism will appropriately signal the scarcity of nat-ural resources, giving incentives to develop improved extraction techniques, explore new reserves and substitute to alternative energy sources. Pessimists fear that these factors

1

In the fifteenth century, the use of coal for domestic heating was banned in London to reduce air pollution. By 1615, the paucity of wood had induced a 180 degree turn and it was wood that was banned for this very purpose.

(12)

Chapter 1: Introduction

will not be able to compensate for the permanent loss of cheaply accessible supplies. Over the last few decades, anthropogenic climate change has emerged as one of the biggest global threats. Though the effect of greenhouse gas (GHG) emissions on the global climate, as well as the impacts of higher temperatures and changing weather patterns on human well-being are still uncertain, the consequences of climate change are potentially catastrophic. Fossil fuels are an important contributor to the problem, accounting for 57% of global GHG emissions (IPCC, 2007).

Policymakers across the globe must contend with the two challenges of energy avail-ability and climate change. Scarcity of fossil fuels in the physical sense has not ma-terialized so far: fossil fuel production has not yet peaked, and physically recoverable reserves of coal and unconventional oil and gas are sufficient for the foreseeable future. Economic exhaustion is a more realistic prospect: the deepwater and tar sand oils that are currently being explored are far more costly to extract than reserves in the Middle East. Nonetheless, the recent shale gas boom in the United States and other countries may herald a protracted era of low natural gas prices. But even if these fossil fuels will be more abundantly and cheaply available than renewable alternatives for a long period, continued reliance on them can gravely exacerbate the climate change problem. Climate scientists already advocate a cold turkey style abandonment of fossil fuels (Kharecha and Hansen, 2008).

Limiting climate change and ensuring the continued availability of low-cost energy is complicated by an important market failure. CO2 emissions are a textbook example of a so-called externality: producers and consumers enjoy the full benefit of emitting a unit of carbon, but face only a fraction of the total costs in terms of climate change, because these are shared between all members of current and future generations. This causes individual decision makers to consume more fossil fuel than would be optimal from a social perspective. A carbon price can realign the individual and social objectives: by requiring each emitter to pay a price equal to the burden his emissions impose on contemporaries and future generations, individual decision makers will only emit a unit of carbon if their private benefit exceeds the social cost of emissions.

Regulators that want to impose such a carbon price face limited spatial and tempo-ral jurisdiction however. Free-riding problems make it difficult to form a global climate coalition: each country would like all other countries to participate in an agreement

(13)

Introduction

so that it can benefit from their abatement efforts in terms of lower temperatures, but continue under business-as-usual itself (Barrett, 1994). An international treaty offers no panacea: since a supranational authority that can audit emission levels and punish coun-tries that do not abide by their commitments would be at odds with national sovereignty, participating countries still have an incentive to shirk.2

The free-rider problem applies even when all agents in the economy are equally con-cerned about climate change. This dissertation instead focuses on the implications for climate and energy policy when players have different objectives. Fossil fuel produc-ing countries care about export revenues as well as climate change and energy scarcity. Saudi Arabia has been accused of deliberately obstructing climate negotiations (De-pledge, 2008), and Canada withdrew from the Kyoto Protocol in 2011. In turn, importers may use environmental policy to capture a share of the fossil fuel rents or for other strate-gic reasons. European proposals to tax emissions embodied in imports3 _{or impose an}

aviation tax4 _{have been criticized by developing countries as ’green protectionism’. The}

inefficiencies that can result from conflicting objectives and strategic behaviour cannot easily be remedied through multilateral bargaining,5 _{because the bargaining outcome}

may not be enforceable for similar reasons as international environmental agreements. Aside from different objectives between countries, there are also conflicts between successive regulators or generations. Each regulator may be committed to long-term emission reductions, but prefer the costs of these reductions to incur after her tenure or lifetime. Examples from multilateral declarations abound. The Rio Declaration from 1992 calls upon states to ”cooperate in a spirit of global partnership to conserve, protect and restore the health and integrity of the Earth’s ecosystem” and ”reduce and eliminate unsustainable patterns of production and consumption”, yet global carbon emissions from energy increased 48% in the next twenty years.6 _{The EU has resolved in 2007 to}

reduce emissions by 80% in 2050 compared to 1990, but the largest and most costly cuts

2

See Di Maria et al. (2013) for a non-technical overview of the channels and magnitude of carbon leakage, i.e. the extent to which emissions increase in non-regulated countries following a unilateral emission reduction by a coalition of concerned countries.

3

India urges rich not to use ”green” protectionism, http://www.reuters.com/article/2009/04/ 07/us-climate-india-bonn-idUSTRE5365FJ20090407, accessed February 11, 2013.

4

EU’s carbon trade plan for aviation is green protectionism, http://news.xinhuanet.com/english/ indepth/2011-12/22/c_131321709.htm, accessed February 11, 2013.

5

In this vein, Harstad (2012) proposes that climate-conscious countries buy the physical coal deposits in less concerned countries, with the intention of leaving them unexploited.

6

Global carbon emissions rise is far bigger than previous estimates, http://www.guardian.co.uk/ environment/2012/jun/21/global-carbon-emissions-record, accessed February 11, 2013.

(14)

are reserved for the latter part of this horizon.

The success of energy or climate policy in achieving its aims, whether securing a sta-ble energy supply or reducing cumulative carbon emissions, thus depends on the actions of fossil fuel suppliers and future generations that may not have the same preferences as the policy maker. Effective regulation must take these reactions into account. After highlighting the importance of an abundant energy supply in Chapter 2, this disserta-tion discusses some implicadisserta-tions of conflicting objectives and strategic consideradisserta-tions for energy and climate policy.

Chapter 2 quantifies the effect of energy abundance on industry location. Attracting jobs and profitable industries is a key concern of policy makers, so it is useful to under-stand what drives firms’ decisions to locate in one jurisdiction or another. The chapter specifically asks whether energy-intensive manufacturing sectors have a tendency to lo-cate in US states that have abundant coal, natural gas, oil and hydro reserves. These energy carriers are traded internationally in well-functioning markets but even within the US, local energy prices can differ by a factor three or more. Especially coal and electricity are costly to transport across large distances, and local regulation further contributes to price disparities across states.

Relatively homogeneous jurisdictions such as US states or OECD countries differ more in their energy endowments than in their endowments of capital and skilled labour, which are traditionally considered as the most important factor endowments for firms’ location decisions. At the same time, manufacturing sectors are more similar in their capital- and skilled labour requirements than in their energy requirements. The empirical results in Chapter 2 confirm that energy abundance is more important for energy-intensive firms than capital abundance is for capital-intensive firms. States with high coal reserves, such as Montana and Wyoming, mainly attract energy-intensive industries because they have lower electricity prices. Natural gas, oil and hydro endowments also have a direct effect on industry location conditional on energy prices, for example because energy-intensive sectors (e.g. Aluminum and Iron and steel mills) sell part of their output to energy extraction firms, and want to locate close to energy reserves for that reason.

Chapters 3 and 4 deal with incentives of fossil fuel suppliers and their implications for climate and energy policy. In most economic markets, suppliers’ decisions how much to produce are more or less separated over time. When firms expect demand to increase in

(15)

Introduction

say ten years, they may start investing in additional capacity ahead of time, but today’s supply need not adjust. Fossil fuels are exhaustible commodities however: each unit that is extracted today cannot be used to satisfy demand in the future. As a result, their owners constantly compare whether it is more profitable to sell their resources today or in the future. They prefer to sell an additional unit today if the profit from selling an extra unit today (the marginal profit) plus the interest on the proceeds exceeds the marginal profit tomorrow, and vice versa. Along the fossil fuel owners’ preferred supply path, marginal profits increase at the interest rate.7

Carbon taxes and investments in renewable alternatives typically reduce fossil fuel demand in the medium- and long term more than in the short run. Carbon policies, like the EU Emissions Trading Scheme, often become stricter over time to give producers and consumers some time to adjust their investment decisions, and R&D efforts only result in lower renewable energy prices with a time lag. When climate policies make selling fossil fuels today comparatively more attractive vis a vis selling them in the future, the policies may accelerate fossil fuel extraction. This observation has sparked the fear that anticipated climate policies will worsen rather than solve the climate problem - a ’green paradox’ (Sinn, 2008a).8

Chapter 3 qualifies this fear by arguing that anticipated carbon taxes are unlikely to increase today’s supply of coal and unconventional oil, which are the most potent threat to the global climate. Because these resources are so abundantly available, the tradeoff between extracting a unit today or in the future is less salient than for low-cost oil and natural gas, which are in much more limited supply. Today’s supply of coal and unconventional oil primarily depends on today’s prices and extraction costs, rather than on expected future market conditions. The chapter derives conditions for which anticipated policies reduce current as well as future emissions. Calibrations suggest that anticipated climate policies are likely to reduce future emissions from coal and unconventional oil much more substantially than they accelerate oil and gas extraction. Chapter 4 abstracts from climate change and focuses on consumer and producer sur-plus in the energy market. An importer (for example the OECD) has the ability to

7

A simple intuition is that a barrel of oil in the ground can be thought of as an asset, which must earn the same return as all other assets in the economy - the interest rate.

8

Gerlagh and Michielsen (2013) summarize the insights from the economic literature on the green paradox.

(16)

develop a substitute for oil by paying an upfront cost, and is both interested in alleviat-ing the physical scarcity of oil and maximizalleviat-ing its share of the oil rent. A monopolistic exporter (say the OPEC) wants to maximize profits while discouraging the OECD from developing the substitute. Because today’s oil supply depends on the OPEC’s expecta-tions about future condiexpecta-tions, the OECD would like to make a binding promise about the innovation time in order to favourably influence the OPEC’s supply schedule. Such a promise would not be credible however, because the OECD has an incentive to renege on its announcement when the oil becomes scarce.

The chapter uses game theoretic methods to derive the OECD’s optimal innovation policy and the OPEC’s optimal supply rule when neither player can commit to future actions. In equilibrium, the OPEC induces the OECD to delay innovation until the oil is exhausted. By innovating earlier, the OECD loses an important benefit of oil consumption, namely delaying the moment at which the substitute’s development costs have to be incurred. Early innovation also causes the OPEC to sell a larger share of its oil reserves just below the substitute price, which is wasteful from the point of view of the OECD. From the OPEC’s perspective, the OECD’s ability to develop a substitute is equivalent to an already available substitute that is more costly to produce. The chapter raises the possibility that two commonly cited objectives of research in renewable energy - securing a sustainable energy source and obtaining better prices from fossil fuel producers9 _{- may actually be at odds with each other.}

Chapter 5 explores the implications of conflicting objectives between generations: how can the current generation best adjust its environmental policy when future generations will make a different tradeoff between consumption and preventing an environmental catastrophe, such as severe climate change or large biodiversity loss? Such a disagreement arises naturally when generations share a common concern for preventing a catastrophe, but attach more weight to their own consumption than to the consumption of their descendants. Each generation would then like to implement an austere environmental policy in the future, but enjoy a comparatively high level of consumption today. When today’s policy makers realize that their successors will not play their part in such a

9

US Senator Maria Cantwell for example argued that ”[t]ransportation fuel choice could [..] reduce the $200 billion ’monopoly premium’ the Department of Energy calculates U.S. consumers currently pay to OPEC and other foreign oil producers each year through excessive petroleum prices.” http: //www.cantwell.senate.gov/news/record.cfm?id=334142, accessed February 12, 2013.

(17)

Introduction

’pollute now, clean up later’ plan, the policy that best enacts the current generation’s preferences depends on the characteristics of the environmental problem.

Firstly, the chapter considers environmental problems that are caused by a sufficiently scarce pollutant, for example local pollution related to the extraction of an exhaustible resource. In this case, today’s generation has an incentive to be conservationist when its descendants have full discretion (in subgame-perfect equilibrium) that is not present when it can fully commit all current and future resource use. Under discretion, future consumption is too rapacious from the current generation’s perspective. By reducing its own consumption, the current generation ensures that the resource stock is consumed more smoothly over time, allowing the ecosystem’s natural recovery to reduce the prob-ability of a catastrophe.

If the environmental risk is expected to recede in the near future, for example be-cause technological change will make the polluting resource obsolete, today’s generation may in contrast have a strategic motive to increase its consumption in subgame-perfect equilibrium. Because the number of future generations that can affect the catastrophe risk is small, the current generation has a direct influence on future decisions. When an increase in today’s consumption causes future generations to become more precaution-ary, today’s generation has an incentive to increase its resource use compared to when it has full commitment power, even if this increases the probability of a catastrophe.

Lastly, the chapter analyzes the case in which the polluting resource is abundant and expected to remain essential for a long period. This model has some relevance for climate change if we do not find a substitute for coal and unconventional oil. Here, the catastrophe becomes a self-fulfilling prophecy. Early generations realize that far-future generations will not respect stringent emission ceilings, and that the pollution stock will reach dangerous levels regardless of their actions. Because any mitigation efforts will be undone by future generations, it becomes optimal to continue under business-as-usual. Though each generation has an explicit desire to prevent a catastrophe, generations may act in equilibrium as if they are indifferent about the catastrophe risk. The chapter shows that intergenerational preferences that explicitly value the long-run future can still result in environmental degradation.

Chapter 6 considers a different type of time-inconsistent preferences than Chapter 5, namely hyperbolic discounting: each generation values its own well-being very highly

(18)

vis a vis their children’s well-being, but does not make as sharp a distinction between their children’s and their grandchildren’s well-being. At the level of an individual, such preferences can explain why people prefer receiving e 200 in one year and one month to e_{150 in one year, but simultaneously favour e 150 today over e 200 in one month, as} well as various kinds of behaviours such as procrastination, addiction, undersaving and lack of exercise. In an intergenerational context, hyperbolic preferences can resolve the tension between high short-term interest rates, which suggest a high degree of impatience, and concerns for far-future generations from stated preference studies and introspection. The positive and normative appeal of these preferences is similar to the preferences in Chapter 5, and future research can shed light which ones are most appropriate for long-lived environmental problems.

The chapter studies the management of a nonrenewable resource with amenity value, for example biodiversity, which provides a range of ecosystem services and is valuable for pharmaceutical research, or the carbon concentration in the atmosphere, which con-tributes to a hospitable climate. The natural resource provides a stream of benefits when left intact, but can also be irreversibly depleted for immediate economic gain, for example by cutting down the habitat of an endangered species for timber production. Today’s generation values the resource’s ability to provide amenity value into the far future relatively highly vis a vis the consumption of their immediate descendants. As a result, today’s generation may consume the resource if it believes that its descendants will do so otherwise, even if it would prefer to see the resource preserved indefinitely. Today’s generation is more inclined to follow a conservationist policy if it is confident that future generations will follow suit.

By contrast, naive policies that ignore the current generation’s inability to rule from the grave are less likely to degrade the environment: naive policy makers do not contend with the possibility of future depletion, and thus have no motive to capitalize the resource before their descendants will. The chapter shows that such an unawareness of future preferences need no longer be a blessing when the resource can regenerate naturally. In this case, naive policies can lead to rapacious depletion in the mistaken belief that future generations will restrain themselves to replenish the resource. The chapter provides an example in which a more resilient ecosystem leads to lower welfare if the ecosystem is managed by naive regulators.

(19)

Chapter 2 The distribution of

energy-intensive sectors in the

US

10 2.1. Introduction

What drives the location of industries? This paper argues that energy is a major deter-minant. Though coal, natural gas and oil are traded internationally at well-established prices, availability and end-user prices differ substantially across regions. Chemical and metal sectors spend 5-15% of their turnover on energy inputs and benefit greatly from being close to energy reserves, in order to minimize costly transport and to take advan-tage of energy subsidies. Particularly for coal, transport costs are high compared to the value at the mine,11 _{and electricity grids are not designed to handle large volumes of}

interregional traffic. At the same time, sought-after energy reserves are highly concen-trated. The Powder River Basin for example contains more than 40% of coal reserves in the US, the world’s most coal-abundant country.

When a select group of industries has a strong incentive to locate near a handful of reserves, variations in energy endowments are an important driver of regional specializa-tion. I test this hypothesis using data on 86 4-digit manufacturing sectors in 50 US states. Coal, natural gas, and to a lesser extent hydro endowments, attract energy-intensive in-dustries. A one standard deviation increase in per capita coal or natural gas endowments increases value added in industries that are more energy-intensive than average by more than 20%. Natural gas, hydro and oil abundance also affect industry location directly

10

This chapter will also appear as Michielsen (2013a). 11

Gerking and Hamilton (2008)

(20)

Chapter 2: Energy-intensive sectors in the US

once I condition on (instrumented) energy prices, possibly through forward and backward linkages between extractive industries and energy-intensive manufacturing.

The location of manufacturing industries is of great interest to policy makers, and commentators repeatedly highlight the importance of reliable and affordable energy avail-ability for energy-intensive sectors. After the German government’s decision to abandon nuclear power in 2011, a leading national newspaper wrote: ”What will the new energy age cost us [..] in terms of money and jobs? [..] Energy is the lifeblood of industry, which in turn is the basis of our economy and our prosperity. A stable energy supply is taken for granted [..] and is an enormously important locational advantage when attracting foreign investment.”12 _{This paper contributes to quantifying the locational advantage}

that energy abundance provides.

Since the development of the Heckscher-Ohlin model, many studies have tested the factor abundance hypothesis, which states that regions specialize in industries that use their abundant factors.13 _{These studies mostly focus on capital and (skilled) labour}

however.14 _{Though capital is a stock input and energy a flow, expenditures on capital}

and energy as a fraction of turnover (3% and 2%, respectively)15 _{suggest that they are}

an equally important determinant of industry location. Sectors are more similar in their capital- and skilled labour requirements than in their energy requirements, and capital and skilled labour are distributed more evenly across regions than energy reserves. I find that capital abundance is less important for capital-intensive industries than energy abundance is for energy-intensive industries.

To the best of my knowledge, this paper is the first to consider in detail the relation between energy abundance and industry location at the regional level. By focusing on US states instead of countries, I reduce the risk that heterogeneity in technology and consumer tastes dilutes the relation between factor endowments and specialization. De-tailed data on energy reserves allow me to insulate effects for four endowments: coal, natural gas, oil and hydro. Furthermore, my primary interest in energy reserves makes

12

Die Welt, May 30th 2011 (translation by Der Spiegel) 13

c.f. Bowen et al. (1987), Trefler (1995), Davis et al. (1997) and Romalis (2004) 14

I list three important exceptions. Hillman and Bullard (1978) assume capital-energy complementar-ity and conjecture that energy is a source of comparative disadvantage for the US. Ellison and Glaeser (1999) find that US states with low energy prices have a higher activity in energy-intensive sectors, but the direction of causality is unclear. Closely related to the current paper, Gerlagh and Mathys (2011) use data on 14 OECD countries and find that energy-abundant countries produce and export more in energy-intensive industries.

15

Average during 2001-2009 in US manufacturing, Economic Census

(21)

Methodology

interpretation of the results less prone to endogeneity in the distribution of production factors across states (Schott, 2003). The rest of this paper is organized as follows. Sec-tion 2.2 outlines the methodology. SecSec-tion 2.3 gives an overview of the data. SecSec-tion 2.4 presents the results, and section 2.5 concludes.

2.2. Methodology

The three main theories of industry location are Ricardo’s theory of comparative advan-tage, the factor abundance hypothesis and the new economic geography literature which emphasizes increasing returns and external economies. While I am foremostly interested in testing whether the factor abundance hypothesis hold for energy carriers, I control for explanations given by the other two theories. This paper is closely related method-ologically to Midelfart-Knarvik et al. (2000), Crafts and Mulatu (2005) and Gerlagh and Mathys (2011). In their report on industry location in the EU, Midelfart-Knarvik et al. (2000) interact regional characteristics with sectoral characteristics. If region i has a desirable characteristic j, say an abundance of capital, all sectors will be interested in locating in region i. However, a region’s capacity to absorb industries is bounded. The industries that end up in region i are the ones that benefit most from capital abundance and low capital prices, i.e. capital-intensive sectors. Capital-extensive sectors locate somewhere else and are thus underrepresented in region i.

This approach can be applied to production factors, as in the example above, as well as new economic geography (NEG) effects. I include three of these. Industries that have strong forward or backward linkages may agglomerate near large markets, to be close to other producers. I control for this by interacting regional market potential with sectoral forward and backward linkages. Furthermore, industries with large economies of scale may locate in central locations to minimize transport costs. I capture this by interacting market potential with average plant size. I include state-year and sector-year fixed effects.16 _{The state-year fixed effects control for any changes in state characteristics}

16

Midelfart-Knarvik et al. (2000) include cutoff levels for state endowments and sectoral intensities in the interaction terms. The interpretation of an endowment cutoff for skilled labour is the endowment level such that an industry’s activity does not depend on the skilled labour intensity of the industry. Analogously, the skilled labour intensity cutoff signals the intensity for which industries do not consider the state endowment of skilled labour when making their location decision. Mulatu et al. (2010), studying the effects of environmental regulation on European industry location, present and discuss

(22)

Table 2.1: Nomenclature i state subscript

s sector subscript t time subscript

j production factor subscript l economic geography subscript E set of energy production factors N set of non-energy production factors V Ai,s,t value added

πj,s,t sectoral factor intensities

θj,i,t state factor endowments

pj,i,t state energy prices

ξj,i state deregulation indicators

σl,s sectoral economic geography characteristics

χi,t state market potential

that affect all sectors, which may include changes in the tax code or labour regulation. The sector-year fixed effects absorb any unobserved nationwide sectoral trends, such as price changes for crucial inputs or changes in consumer tastes. I omit sector-state fixed effects as energy reserves and sectoral energy intensities, which enter into the interaction effects of interest, do not vary much over time. To control for this persistence, I cluster error terms by sector-state pair.

I measure industrial activity by value added.1718 _{Table 2.1 describes the notation. I}

estimate industry location as dependent on a set of production factor interaction terms, three economic geography interaction terms and control variables. The equation I esti-mate is ln V Ai,s,t = αi,t + βs,t+ X j∈{E∪N } γjπj,s,tθj,i,t+ X l δlσl,sχi,t + ǫi,s,t (2.1)

their estimated cutoff levels in some detail. Because I include fixed effects, I cannot identify these cutoffs.

17

Midelfart-Knarvik et al. (2000) normalize the left-hand variable by country and sector size. In my specification, size effects are absorbed by the fixed effects.

18

In a robustness check in the Appendix, I use the log of employment as left-hand variable.

(23)

Methodology

The factor abundance hypothesis predicts that the coefficients on the factor interac-tion terms γj are positive, indicating that energy-intensive industries have a higher value

added in energy-abundant states. New economic geography posits that the δlcoefficients

are positive. For ease of interpretation, I discretize the sectoral characteristics such that they are equal to one (zero) if they are higher (lower) than average.19 _{As in Mulatu et al.}

(2010), I normalize state characteristics by their standard deviation. The γj coefficients

in (2.1) are then comparable across factors. The magnitude of the coefficient on the in-teraction term for e.g. capital tells us how much variations in capital endowments across states affect the location of industries that are more capital-intensive than average.

I then decompose the effect of energy abundance on the location of energy-intensive industries into a direct effect and an indirect effect through energy prices, as predicted by factor endowment models of trade (c.f. Romalis (2004)). Energy prices are poten-tially endogenous to industry location, so I instrument for prices using a two-step GMM approach. ln V Ai,s,t = ˜αi,t + ˜βs,t+ X j∈{E∪N } ˜ γjπj,s,tθj,i,t+ X j∈{E} ˜ ζjπj,s,tpj,i,t+ X l ˜ δlσl,sχi,t+ ˜ǫi,s,t (2.2) For capital and skilled labour, I preserve the sectoral intensity × state endowment in-teraction term. For energy, I interact sectoral intensities both with state endowments and state prices. ˜γj measures the direct effect of energy abundance on energy-intensive

sectors, such as forward and backward linkages between the extractive activities and manufacturing sectors (Michaels, 2010), and effects on economic fundamentals such as institutions and infrastructure (c.f. Papyrakis and Gerlagh (2007)). Using the same nor-malization as for the other interaction terms, the interpretation of ˜ζj is the percentage

change in industrial activity in energy-intensive sectors if energy prices increase by one standard deviation.

For electricity and natural gas, I instrument the price interaction terms πj,s,tpj,i,t

with deregulation interaction terms πj,s,tξj,i. For electricity, I use an indicator whether

the electricity sector was deregulated at the start of my sample period.20 _{A large part}

of electricity price differences across states in the eighties and nineties was not related

19

In the Appendix, I present robustness checks with continuous factor intensities. 20

Source: EIA (2000).

(24)

to fundamentals such as energy abundance and utilization, but to inefficient generation investments and long-term contracts between generators and utilities that stipulated high prices (Joskow, 2000; Borenstein and Bushnell, 2000). Deregulation took place in states in which the gap between regulated prices and the market value of electricity was largest, primarily on the West Coast and in the Northeast. The restructuring proved to be unsuccessful in most states, and did little to curb prices (Blumsack et al., 2006). The deregulation indicator is thus a proxy for long-standing inefficiencies in the electricity sector that, unlike energy reserves, only affect industry location through electricity prices. Because the deregulating states do not have large energy-intensive manufacturing sectors, reverse causality from the composition of the manufacturing industry to deregulation is unlikely.

For natural gas, I employ two similar instruments: indicator variables whether a state adopted a price cap or cost incentive measure for natural gas utilities before the start of my sample period.21 _{The utilities, which enjoy natural monopolies, were traditionally}

subject to rate-of-return regulation. In the nineties and the beginning of the aughts, a number of states introduced price caps and cost-incentive measures. While the two measures are quite different in nature, Hlasny (2011) demonstrates that they were both most likely to be implemented in states with high natural gas prices conditional on geographic characteristics such as natural gas endowments and climate, and in states with high concentration ratios for natural gas distribution. Like in the case of electricity, the reforms did not reduce consumer prices (Hlasny, 2006). Furthermore, I use the population-weighted number of heating degree days22 _{as an instrument for natural gas}

and fuel oil prices. These energy types are widely used as a heating fuel, and temperature-induced variations in demand for heating across states will affect natural gas and fuel oil prices across states.

2.3. Data

I use state-level US panel data containing information on energy reserves, sectoral output and factor inputs, covering a period of 2001-2009. Appendix 2.A.1 contains a full list

21

Source: Hlasny (2008). 22

Source: National Oceanic and Atmospheric Administration.

(25)

Data

of data sources and definitions. I compute energy abundance as proven reserves per capita. I assume that proven reserves are independent of industrial activity. Though per-capita measures are potentially endogenous, the bias goes against my hypothesis: if energy-abundant states attract more people because they offer better employment opportunities, energy reserves per capita are similar across states. Variations in reserves per capita then have a smaller influence on the location of energy-intensive industries than if reserves per capita are exogenous.

Moreover, area normalizations are distortive as land area is an imperfect indicator of economic potential. Montana and Wyoming, the states with the first and third largest coal reserves, are considerably larger than Illinois, which has the second largest reserves. Comparing these three states using an area-normalized measure of coal abundance, we would predict Illinois to have a comparative advantage in energy-intensive sectors and Montana and Wyoming in energy-extensive, i.e. labour-intensive, sectors. This pre-diction is implausible as Illinois is more than ten times as populous as the latter two states. Appendix 2.A.2 contains a robustness check in which I define energy abundance as reserves per square mile of land area.

I measure energy intensity as national energy expenditures per employee. The re-sults do not change when measuring energy intensity as energy expenditures per dollar of value added, to ensure that my measure of energy intensity is not clouded by varia-tions in labour intensity. The Energy Information Agency (EIA) has data on four energy endowments: natural gas, coal, oil and hydro.23 _{The left-hand variable and factor}

in-tensities come from the Annual Survey of Manufactures (ASM) and are observed at the 4-digit NAICS level. For energy, I have data on electricity- and fuel intensity. I use a perpetual inventory approach to construct capital endowments and intensities.

Table 2.2 shows the least and most energy-intensive sectors. Industries that process raw materials, as well as chemical industries, tend to require a lot of energy. Sectors that are further down the production chain, especially those that cater to consumers, are typically energy extensive. Figure 2.1 depicts the sources of electricity generation in in the US. Coal is the predominant source of electricity, accounting for nearly 50%

23

Hydropower generation is constrained by geographic characteristics, so I regard hydropower capacity as an adequate proxy for the endowment of suitable hydropower generation locations. The location of nuclear power plants may be endogenous, so I do not include this electricity source in the econometric analysis.

(26)

C h ap te r 2: E n er gy -i n te n si ve se ct or s in th e US

Table 2.2: Four-digit NAICS sectors with lowest and highest shares of turnover spent on energy

Least energy intensive Energy expenditures_Turnover Most energy intensive Energy expenditures_Turnover

3341 Computers & peripheral equipment 0.30% 3315 Foundries 4.77%

3122 Tobacco 0.30% 3252 Resin, syn rubber, & artificial syn fibers 4.83%

3361 Motor vehicles 0.39% 3279 Other nonmetallic mineral products 4.85%

3342 Communications equipment 0.45% 3313 Aluminum 7.10%

3343 Audio and video equipment 0.46% 3271 Clay products & refractories 7.26%

3379 Other furniture 0.48% 3251 Basic chemicals 7.31%

3369 Other transportation equipment 0.51% 3311 Iron & steel mills 7.42%

3345 Electronic instruments 0.59% 3272 Glass & glass products 7.62%

3391 Medical equipment & supplies 0.60% 3221 Pulp, paper, & paperboard mills 9.03%

3362 Motor vehicle bodies & trailers 0.60% 3274 Lime and gypsum 13.91%

Shares are averaged over 2001-2009, Annual Survey of Manufactures

(27)

Data

Table 2.3: Sample correlation coefficients for energy interaction terms

(1) (2) (3) (4) (5)

(1) electricity intensity × coal abundance 1 (2) electricity intensity × natural gas abundance 0.33 1 (3) electricity intensity × hydro abundance 0.24 0.11 1 (4) fuel intensity × natural gas abundance 0.26 0.81 0.07 1 (5) fuel intensity × oil abundance 0.40 0.62 0.14 0.75 1

of total generation. As natural gas is also an important input, I interact natural gas endowments with electricity intensity as well as with fuel intensity.

Table 2.3 presents the correlation between the energy interaction terms. Multi-collinearity between the natural gas and oil interaction terms makes it difficult to disen-tangle the effect of natural gas and oil endowments on the location of energy-intensive industries. In the Appendix, I use a more disaggregated measure of energy intensity, decomposing fuel intensity into natural gas-, distillate fuel oil- and residual fuel oil in-tensity. The extra information on energy intensities comes at the cost of observing 3-digit rather than 4-digit sectors.

Coal 48.5%

Natural Gas 21.6%

Nuclear 19.4%

Hydroelectric 6.0% Other Renewables 2.5%Other 2.0%

Figure 2.1: Electricity generation in the US by source in 2007 (EIA)

Lastly, Table 2.4 shows the correlation between energy reserves and prices. In accor-dance with the factor abunaccor-dance hypothesis, industrial energy prices tend to be lower in states with large energy reserves. Consistent with Figure 2.1, coal and hydro abundance are negatively correlated with electricity prices. The relation between natural gas

(28)

Table 2.4: Correlation between state energy abundance (rows) and prices (columns) in 2007

Electricity Natural gas Distillate fuel oil Residual fuel oil

Coal _−0.26∗ _{−0.37 ∗ ∗∗} 0.18 _{−0.33 ∗ ∗}

Hydro _−0.25∗ _−0.04 0.04 _−0.20

Natural gas _−0.05 _{−0.53 ∗ ∗∗} 0.00 _{−0.37 ∗ ∗∗}

Oil _−0.03 _{−0.53 ∗ ∗∗} 0.06 _{−0.39 ∗ ∗∗}

a

Prices are for the industrial sector. Abundance and prices are truncated at the 95th percentile. Asterisks denote significance at the 10% (*), 5% (**) and 1% (***) level. Source: State Energy Data System, EIA.

dance and electricity prices is weak however. Natural gas and oil abundance are strongly negatively correlated with natural gas and residual fuel oil prices respectively. There is no significant relation between distillate fuel oil prices and oil reserves, suggesting that the distillate fuel oil market is more nationally integrated than the residual fuel oil one.

2.4. Results

Table 2.5 presents the results of regression (2.1). The effect of energy endowments on industry location is both statistically and economically significant. A one standard deviation increase in per capita coal or natural gas endowments increases the value added of electricity-intensive sectors by 23%. The effect for hydro is slightly lower, as hydropower constitutes a smaller share of total electricity generation. Natural gas endowments play an even stronger role in the location of fuel-intensive industries: a one standard deviation increase in endowments brings about a 39% increase in value added in fuel-intensive sectors. I find no evidence that oil endowments matter for the location of fuel-intensive industries. The senstivity analyses with finer energy-intensity disaggregation in the Appendix suggest that the negative sign on the oil interaction term is caused by multicollinearity with the natural gas interaction terms. The US possess 4.5% of world conventional natural gas reserves, but only 1.5% of oil reserves.24 _For

oil-intensive industries, access to imports may be more important than for natural gas-intensive industries.

24

International Energy Statistics 2009, EIA

(29)

R

es

u

lt

s

Table 2.5: Location of 4-digit NAICS sectors

(1) (2) (3)

electricity intensity × coal abundance _(0.058)***0.23 _(0.061)***0.25

electricity intensity × natural gas abundance _(0.069)***0.25 −0.04_(0.074)

electricity intensity × hydro abundance _(0.047)***0.17 _(0.046)***0.18

fuel intensity × natural gas abundance _(0.076)***0.39 _(0.088)***0.48

fuel intensity × oil abundance −0.05_(0.104) −0.16_(0.112)

capital intensity × capital abundance _(0.045)0.07 _(0.045)0.07 _(0.045)0.07

skill intensity × skill abundance _(0.030)*0.05 _(0.030)*0.05 _(0.030)*0.06

forward linkages × market potential _(0.036)0.05 _(0.035)0.01 _(0.036)0.05

backward linkages × market potential −0.12_(0.035)*** −0.13_(0.035)*** −0.12_(0.035)***

scale economies × market potential −0.07_(0.039)* −0.08_(0.039)** −0.08_(0.039)**

Number of observations 18423 18908 18423

Number of clusters 2971 3071 2971

Adjusted R2 _0.57 _0.57 _0.57

This table reports coefficient estimates for regression equation (2.1). Variable definitions are given in Tables 2.7 and 2.8. Two-way fixed effects (state-year and sector-year) are included. Error terms are clustered by sector-state pair. Standard errors in parentheses. Asterisks denote significance at the 10% (*), 5% (**) and 1% (***) level. Sector characteristics are equal to one if they are larger than average, and zero if they are smaller than average. State characteristics are divided by their standard deviation.

(30)

Figure 2.2: Coal endowments per capita (left) and value added in electricity-intensive industries per capita (right), 2007

Figure 2.3: Natural gas endowments per capita (left) and value added in fuel-intensive industries per capita (right), 2007

Figure 2.2 illustrates that electricity-intensive industries tend to locate in states with large coal endowments. The left panel shows coal endowments per capita; the right panel value added in electricity-intensive sectors per capita. There are two main coal producing regions in the US: the Appalachians (with Illinois, Kentucky and West Virginia as the most abundant states) and the Western Coal Region (with large reserves in Wyoming, North Dakota and Montana). With the exception of North Dakota, states in these regions also have a high value added in electricity-intensive sectors. Figure 2.3 shows that fuel-intensive sectors are overrepresented in natural-gas abundant Great Plains (Wyoming, Oklahoma and Colorado) and Gulf Coast (Texas and Louisiana) states.

By comparison, the distribution of capital and skilled labour across the US has a much smaller influence on the location of industries that rely on these factors more than average. The coefficients (0.07 and 0.05 respectively) are smaller in absolute value than those of the energy interaction terms, and not significant at the 5% level. Where does

(31)

Results .63 1.25 2.5 5 10 20 40 60

Energy expenditures per employee in 1983 USD x 1000

10 20 40

Average wage per employee in 1983 USD x 1000

.63 1.25 2.5 5 10 20 40 60

Energy expenditures per employee in 1983 USD x 1000

25 50 100 200 400 600 Capital stock per employee in 1983 USD x 1000

Figure 2.4: Labour, capital and energy intensities in 4-digit US manufacturing sectors. (Source: Economic Census, 2007)

this difference come from? Firstly, the variation in energy intensities across sectors is much higher than the variation in capital- and skill intensities (Figure 2.4). Energy expenditures in the most energy-intensive sector are 152 times as large as in the least energy-intensive sector. The largest difference factors for capital- and skill intensities are 22.8 and 3.3, respectively. For industries that are extremely energy-intensive, being close to energy reserves can be a crucial consideration in the location decision. Because capital- and skill-intensities are much less skewed, locating in capital- or skill-abundant states is not as overriding a concern for capital- and skill-intensive industries.

Secondly, state energy endowments are much more concentrated than capital and skilled labour endowments. Figure 2.5 illustrates this point for natural gas reserves. More than half of states has no natural gas at all, whereas the most abundant state (Wyoming) has 155 times as many reserves per capita as New York. Sectors that require a lot of energy thus have limited options if they want to locate close to energy reserves. Skilled labour and physical capital on the other hand are available in every state. The fraction of adults with a Bachelor’s degree or higher differs by a factor 2.2 at most; the largest difference in physical capital per capita is a factor 8.1. As capital and skill

(32)

Chapter 2: Energy-intensive sectors in the US AK AL AR CA CO KS KY LA MI MS MT ND NM NY OH OK PA TX UT VA WY 60 40 20 10 5 2.5 1.25 .63

per Capita Dry Natural Gas Proved Reserves in 1000 Cubic Feet

.2 .25 .3 .35 .4

Fraction of pop. over 25 with a Bachelor’s degree

AK AL AR CA CO KS KY LA MI MS MT ND NM NY OH OK PA TX UT VA WY 60 40 20 10 5 2.5 1.25 .63

per Capita Dry Natural Gas Proved Reserves in 1000 Cubic Feet

2 4 6 8 10

per Capita Capital Stock in 1983 USD x1000

Figure 2.5: Natural gas, capital and skilled labour endowments of US states (Source: EIA and Economic Census, 2007). Dots without labels represent states with little or no gas reserves.

endowments vary much less across states than energy reserves, we can expect variations in endowments to play a smaller role in the location decision for capital- and skill-intensive industries than for energy-intensive industries.

I find no support for the new economic geography hypotheses. Market potential does not significantly affect the location of sectors that have more forward linkages than average. For backward linkages and economies of scale, my findings contradict the NEG prediction: sectors with strong backward linkages and scale economies are under- rather than overrepresented in large markets.

Table 2.6 shows the results when I include both endowment and price interaction terms (for brevity, I do not report the coefficients on the capital, skill and NGE interaction terms). The first-stage IV results are in Table 2.9 in the Appendix. Energy abundance is significant in the first-stage IV regressions. A one standard deviation increase in natural gas abundance is associated with a 38% decrease in natural gas prices. Coal abundance has a stronger influence on electricity prices than hydro or natural gas abundance, which is consistent with the high share of coal-fired electricity generation in Figure 2.1. States

(33)

R

es

u

lt

s

Table 2.6: OLS and GMM-IV coefficient estimates for location of 4-digit NAICS industries with energy intensity and price interaction terms

(1) (2) (3) (4)

OLS GMM-IV OLS GMM-IV OLS GMM-IV OLS GMM-IV

electricity intensity × electricity price _(0.042)***-0.24 _(0.094)***-0.33

fuel intensity × natural gas price _(0.045)***-0.18 _(0.229)**-0.58

fuel intensity × distillate fuel oil price _(0.035)**-0.07 _(0.088)**-0.18

fuel intensity × residual fuel oil price _(0.033)**-0.07 _(225.866)1.63

electricity intensity × coal abundance _(0.060)**0.14 _(0.065)0.08 electricity intensity × hydro abundance _(0.047)***0.13 _(0.049)**0.12 electricity intensity × natural gas abundance _(0.069)***0.22 _(0.063)***0.19

fuel intensity × natural gas abundance _(0.062)***0.28 _(0.161)-0.01

fuel intensity × oil abundance _(0.086)***0.28 _(0.071)***0.24 _(0.089)***0.27 _(67.074)0.80

Number of observations 18423 18423 18908 18908 18908 18908 18908 18908

Number of clusters 2971 2971 3071 3071 3071 3071 3071 3071

Adjusted R2

0.57 0.56 0.57 -0.01 0.56 0.44 0.56 -0.10

Hansen overidentification p-value 1.00

Kleibergen-Paap statistic 237.48 19.67 237.79 21.90

This table reports coefficient estimates for regression equation (2.2). All regressions include capital, skill and NGE interaction terms. In the GMM-IV regressions, the energy price interaction terms are instrumented. The excluded instruments are the following. In (1): electricity intensity × electricity deregulation. In (2): fuel intensity× natural gas price cap, fuel intensity× natural gas cost incentive and fuel intensity× heating degree days. In (3) and (4): fuel intensity × heating degree days. Results for the first-stage IV regressions are reported in Table 2.9. Two-way fixed effects (state-year and sector-year) are included. Error terms are clustered by sector-state pair. Standard errors in parentheses. Asterisks denote significance at the 10% (*), 5% (**) and 1% (***) level. Sector characteristics are equal to one if they are larger than average, and zero if they are smaller than average. State characteristics are divided by their standard deviation.

(34)

that adopted deregulation are characterized by higher electricity and natural gas prices. The coefficients on the endowment interaction terms in the second stage are smaller than in specification (2.1) without price interaction terms. The results indicate that part of the effect of energy endowments on the location of energy-intensive industries goes through energy prices. Energy abundance causes lower energy prices, which in turn attract energy-intensive manufacturing sectors.

Natural gas, hydro and oil endowments also have a direct effect on the location of energy-intensive industries; for coal, the direct effect becomes insignificant when I instrument for electricity prices. Compared to coal, the former three energy types are characterized by a more capital-intensive extraction or generation process, which may result in stronger linkages with manufacturing.

When I instrument for energy prices, the coefficients on the price interaction terms are larger in absolute value than in the OLS regressions. In the OLS results, a one standard deviation increase in electricity or natural gas prices results in a 24% or 18% decrease in value added in electricity- or fuel-intensive industries, respectively. In the IV results, the decrease is 33% and 58%. This result suggests that the presence of large energy-intensive sectors drives up energy prices in energy-abundant states, ameliorating the effect of energy endowments on prices. This demand effect is stronger than possible negative influences of energy-intensive industries on energy prices, for example through lobbying.

The larger magnitudes of the IV coefficients for price interaction terms may also be caused by measurement error in the industrial energy prices.25 _{If energy abundance is}

better correlated with the true prices than the observed prices are, the IV estimates will be larger in absolute value than the OLS estimates. The large standard deviations in IV regression 4 suggest that the number of heating degree days is not an informative instrument for residual fuel prices.

Appendix 2.A.2 presents the robustness checks. The effect of hydro abundance on specialization is less robust than that of coal and natural gas, but the main conclusions remain unaltered.

25

Gerlagh and Mathys (2011) propose a method to recover energy prices that is based on the optimality conditions for a Cobb-Douglas production function. They regress observed energy use on a sector- and country fixed effect. The coefficients on the country fixed effects can then be interpreted as a proxy for marginal costs. This approach requires energy use data with variation across both sectors and states, to which I do not have access.

(35)

Conclusion

2.5. Conclusion

Though energy is often overlooked in industry location analyses because it is a tradable commodity, it plays a significant role in the distribution of manufacturing sectors. My findings suggest that energy is more important than capital and skilled labour for the lo-cation of manufacturing industries in the US. Energy abundance affects industry lolo-cation indirectly through lower energy prices, but also directly when I condition on prices. The analysis in this paper focuses on reserves as an exogenous source of energy abundance, but policy makers can also influence energy availability, for example through investments in nuclear, solar and wind energy. Considering the strong influence of coal and hydro endowments on the location of electricity-intensive industries, such investments can play an important role in attracting value added and employment in energy-intensive sectors.

2.A. Appendix

2.A.1

.

Data definitions and sources

Tables 2.7 and 2.8 list the definitions and sources of the state and sector characteristics, respectively. I measure state capital abundance by the manufacturing capital stock per capita and sectoral capital intensity by the capital stock per employee. I construct state and sectoral capital stocks using a perpetual inventory method.26 _{Denote capital stocks}

by K, capital expenditures by I and the geometric decay rate by δ. State and sectoral capital stocks evolve according to

Ki,t = (1 − δ) Ki,t−1+ Ii,t (2.3a)

Ks,t = (1 − δ) Ks,t−1+ Is,t (2.3b)

26

The US Census’ Quarterly Financial Report directly measures capital stocks for 3-digit NAICS sectors, so I do not need a perpetual inventory approach for sectoral capital intensities for the robustness check with disaggregated energy intensities in the Appendix.

(36)

Harberger (1978) notes that when capital growth It/Kt−1− δ equals output growth g,

the initial capital stock can be calculated as

Kt−1= It/ (δ + g) (2.4)

I set δ = 0.035 and g = 0.03. My data on state capital expenditures start in 1987; the data on sectoral expenditures in 1997, because of the transition from the SIC to NAICS classification.

2.A.2

.

Supplementary tables and robustness checks

Table 2.9 presents the first-stage regression results for specification (2.2), which are discussed in the main text. The results indicate that energy prices are lower in energy-abundant states. The first-stage coefficients should be interpreted with care, as they are overproportionally driven by states with large energy-intensive sectors. The coefficient on the coal interaction term implies that a one standard deviation increase in coal abundance is associated with a decrease in electricity prices of 0.45 standard deviations.

Table 2.10 shows the regression coefficients of (2.1) when I normalize energy- and capital abundance by land area instead of population. The coefficients on all interaction terms for energy and capital are smaller in absolute value than in the main specification. The results do not suggest that energy and capital endowments attract migration. The coefficients for coal and hydro decrease more than for natural gas, as area normaliza-tions introduce more noise into the coal and hydro abundance than in the natural gas abundance measure. The variation in population density between states with the largest absolute coal27 _{and hydro}28 _{endowments is larger than between states with the largest}

natural gas29 _endowments.

Table 2.12 lists the results for regression equation (2.1) when I decompose fuel inten-sity into natural gas inteninten-sity, distillate fuel oil inteninten-sity and residual fuel oil inteninten-sity. The correlation between the natural gas and oil interaction terms is much lower than in the main specification, as can be seen in Table 2.11. Natural gas intensity and electricity intensity are perfectly correlated however. The negative coefficient on the oil

interac-27

Montana, Illinois, Wyoming, Kentucky, West Virginia 28

Washington, California, Oregon, New York, Alabama 29

Texas, Wyoming, Oklahoma, Colorado, New Mexico

(37)

Ap

p

en

d

ix

Table 2.7: State Characteristics

Variable Definition Source

Skilled labour abundance Fraction of population over 25 with a Bachelor’s degree or higher US Census Bureau

Capital expenditures Capital expenditures in all manufacturing sectors Annual Survey of Manufactures

Capital abundance Capital stock per capita Eqns (2.3a) and (2.4)

Coal abundance Estimated recoverable coal reserves per capita Energy Information Administration

Natural gas abundance Dry natural gas reserves per capita Energy Information Administration

Oil abundance Crude oil reserves per capita a

Energy Information Administration

Hydro abundance Summer capacity hydroelectric generation per capita Energy Information Administration

Electricity price Electricity price in the industrial sector State Energy Data System, EIA

Natural gas price Natural gas price in the industrial sector State Energy Data System, EIA

Distillate fuel oil price Distillate fuel oil price in the industrial sector State Energy Data System, EIA Residual fuel oil price Residual fuel oil price in the industrial sector State Energy Data System, EIA

Market potential P_i′

population in state i′ _{in 100.000} max distance between states i and i′ _{in miles,}b

100 !

US Census Bureau Electricity deregulation 1 if electricity restructuring legislation enabled by July 2000, 0 o.w. (EIA, 2000) Natural gas price cap 1 if price caps implemented in 2000 or earlier, 0 otherwise (Hlasny, 2008) Natural gas cost incentive 1 if cost incentive measures implemented in 2000 or earlier, 0 otherwise (Hlasny, 2008)

Heating degree days Total heating degree days weighted by population NOAA

All state characteristics are truncated at the 95th percentile and divided by the yearly standard deviation. I therefore provide no units of measurement. Except for the electricity and natural gas deregulation indicators, the dimension of all state characteristics is i, t. Monetary values are adjusted by the consumer price index from the Bureau of Economic Analysis. a

Does not include Federal Offshore Reserves b

I follow Harris (1954). Distances are for the quickest route between the largest cities (2000) in both states (Source: www.mileage-charts.com). Distances to and from Burlington, VT and Honolulu, HI are from Google Maps.

(38)

C h ap te r 2: E n er gy -i n te n si ve se ct or s in th e US

Table 2.8: Sector Characteristics

Variable Dimension Definition Source

Value added i, s, t Value added Annual Survey of Manufactures

Employment i, s, t Number of employees Annual Survey of Manufactures

Skilled labour intensity s, t Nationwide average wage per employee Annual Survey of Manufactures

Capital expenditures s, t Capital expenditures in all manufacturing sectors Annual Survey of Manufactures

Capital intensity s, t Capital stock per employee Eqns (2.3b) and (2.4)

Electricity intensity s, t Quantity of electricity purchased per employee Annual Survey of Manufactures

Fuel intensity s, t Cost of purchased fuels per employee Annual Survey of Manufactures

Forward linkages sa

Row sum of the Ghosh inverse Bureau of Economic Analysis

Backward linkages sa

Column sum of the Leontief inverse Bureau of Economic Analysis

Scale economies s, t Nationwide average establishment size County Business Patterns

Sector characteristics are equal to one (zero) if the sector characteristic is higher (lower) than the yearly average across sectors. a

The input-output matrix is only available for 2002. I equate forward and backward linkages to the 2002 value for the whole sample period.

(39)

Appendix

tion term in the main specification disappears, though the oil interaction terms are not significant when I control for natural gas abundance and intensity.

Table 2.13 presents the results when I use the log of employment, rather than the log of value added, as dependent variable. All interaction terms have slightly smaller coefficients than in the main specification, though the significance levels are unchanged. A possible explanation is that factor prices are lower in states in which the factor is abundant. For a given output level, capital-intensive industries then have lower costs, and hence a higher value added, in capital-abundant states. By contrast, the number of employees per unit of physical output is more likely to be constant across capital-abundant and capital-scarce states.

Table 2.14 contains the coefficient estimates when energy- and capital intensity are defined as energy expenditures and capital stock per dollar of value added, respectively. Like in Table 2.13, the coefficients are slightly lower in absolute value. The distribution of energy expenditures per employee is more skewed than that of energy expenditures per dollar of value added. Therefore, the average energy-intensity using the alternative metric is lower than the average intensity using the main metric, and more industries classify as energy intensive than in the main specification. Since energy endowments are less important in the location decision of the marginal energy-intensive industries, the energy interaction terms lose some explanatory power.

Table 2.15 shows the results when I do not discretize the sector intensities, to ver-ify whether we lose information as a result of the discretization. The first energy in-teraction term, as well as the capital and skill inin-teraction terms, are slightly larger than in the main specification. The electricity intensity × hydro abundance coeffi-cient is no longer significant, suggesting that hydro abundance does not play as large of a role in the location of the most electricity-intensive industries. The correlation coefficient between electricity intensity × natural gas abundance and fuel intensity × natural gas abundance is higher than in the main specification (0.84 instead of 0.81). Due to increased multicollinearity, the coefficients on the natural gas interaction terms in column (3) diverge.

Essays in environmental and resource economics

Essays in Environmental and Resource Economics

Essays in Environmental and Resource Economics

Acknowledgements

Contents

Chapter 1

Introduction

Chapter 2

The distribution of

energy-intensive sectors in the

US

10

2.1. Introduction

2.2. Methodology

2.3. Data

2.4. Results

2.5. Conclusion

2.A. Appendix

.

.